"Ghost work" is the term used by anthropologist Mary L. Gray for the invisible work that fuels our technological platforms. When Gray, a senior researcher at Microsoft Research, first came to the company, she discovered that the creation of artificial intelligence requires people to manage and clean up the data to feed the training algorithms. "Basically, I started asking the engineers and informatics around me:" Who are the people you pay to do this job of tagging images and sorting tasks and cleaning the databases? " 39; "says Gray. Some people said they did not know. Others said they did not want to know and were worried that if they were observed too much they could find unpleasant working conditions.
So Gray decided to find out for herself. Who are people, often invisible, who perform the tasks necessary for these platforms to run? Why do they do this work and why do they leave? What are your working conditions?
Gray ended up collaborating with his main MSR researcher, Siddharth Suri, to write Ghost Work: How to Prevent Silicon Valley from Building a New World Low Class (Houghton Mifflin Harcourt).
The Verge spoke with Gray about the findings of his research and what they mean for the future of employment.
This interview has been slightly edited for clarity.
Labeling data to feed algorithms is an obvious example of phantom work. Content moderation is another. What are other examples?
Completing surveys, subtitling and translation work, any type of transcription service. Realization of web searches, verification of location addresses, beta tests, user tests for user designs. Anything that can be considered as a work of knowledge, such as content creation, editorial writing, design. Whatever The list is endless. All those are tasks that can be distributed online. It's all we're used to seeing in the office, and this is what seems to dismantle it as a full-time job and turn it into projects for thousands of people.
Am I right in thinking that, basically, all technology companies trust or have relied on phantom work?
It would be hard to find a business that would sell itself as AI that does not rely heavily on phantom work to generate its basic product or does not depend on it much today. There are so many new companies and companies out there, anything that calls itself "business ideas" or "intelligence and analysis". That is to use crowdsourcing or collective intelligence, and that is to rely on phantom work. There is no way to avoid the need for people to examine the stacks of what is called unstructured information.
Sometimes, people think that as technology improves, we will no longer need this kind of phantom work. But you write that "the great paradox of AI is that the desire to eliminate human work generates new tasks for humans." So clearly you do not subscribe to that belief. Why not?
What could change are the specific tasks. Believing that AI will never need humans to label data means believing that language will never change, style will never change. Service industries, especially, are so difficult to automate completely because being able to listen to someone's voice and record the silent anger of that person is such a human capacity. So there are cases in which AI, I argue, always falls short.
Engineers are always wonderfully optimistic about opportunities. As an anthropologist, I know how complicated it is to think interculturally about these issues. Even if we get 100 percent of spoken English with a flat Midwestern accent reliably, what happens when you move into the vernacular, the slang and the people who will splice the languages and code change? Every time you see a self-transference of a conversation, you will see the places where the language breaks, often around someone's name.
Those are the types of computational problems that are incredibly difficult to capture for AI because there is not enough data available in a consistent way to model what the next statement will be that someone says they use Spanglish. We have already effectively automated all the easy things.
An interesting thing you mention is that we do not have good labor statistics on the number of people who perform phantom work. Why is that?
The biggest challenge is that the ways in which we count jobs are often related to professional identities, or really defined skills or abilities, and nobody is oriented to a world of work based on projects. We do not have the language to describe an image tag or subtitle. One of the findings of our research is that people have really different mental models. They may or may not identify themselves as self-employed workers. They may or may not identify as journalists if they write for a content farm, and that could change if they decide to answer a question from a survey to help us measure this workforce. Not to mention the fact that phantom work is distributed around the world and there is no global office of labor statistics.
A key question in this book is: who are the people who do the phantom work? So, who are they? It seems they could be almost anyone.
When we recovered our initial set of surveys of the four different platforms we studied, it became clear that there were as many women as there were men, although they worked at different times. The people had university studies, but that was not surprising, since it is related to knowledge work and information services in general.
We are all us. These are the people who, for reasons of social capital, do not have access to a network that would boost them in full-time work. That is the pattern that I see sociologically or anthropologically. They are attending the first generation college more often than not. This is a group of people who do not have strong social ties with the elites.
What are the motivations of people for this work?
There is not a single type of person who does this work or a single motivation. There is a core group of people who are turning to this work, often due to other limitations in their time. People would say that they do not have time to travel to work and that they will be traveling for a job with a similar salary at least two hours, and that would reduce the amount of money they could earn. That is the calculation they are doing here. So they are deciding to resort to this work and effectively. Once they have figured out how to earn enough money on sufficient platforms, they improvise the equivalent of a full-time salary so that they can meet their needs. We call these people "always active" and are converting this into full-time work for the amount of income streams. But this group of people is a small percentage, from 10 to 15 percent, according to the platform. This is what the research tells us about all these platforms. The core group of people is doing most of the work.
Then there are the "regulars", a deep trench of people who can intervene at any time. The usual customers are those who enable people "always active" because if the "always active" is removed, there are enough people in that stack of regular customers who can intervene. They are often caregivers and had other motivations; They were pursuing another passion project or they were going back to education and taking courses, and this gave them a means to finance that.
Finally, there is the long tail of the experimenters, which is the name we gave to the people who try one or two projects, they discover that this is not for them and they leave. The most important part of doing anthropological work is that we could know the people who left and discover why. And it had to do with never connecting with a community of partners to help reduce their costs, feel they do not have enough support, and that this was too difficult to solve. And it was cognitively exhausting.
A characteristic of this type of market is that anyone can work for any other person. What happens in that kind of environment?
For anyone who becomes a regular customer or "always active", they invest and bring the same framework they have for any job. It's an amazing amount of self-control because the workers invest in making sure that the work comes back to the pool. They want to make sure that their partners do well because, if not, that could go against their interests to get the next job.
Companies should invest equally in this responsibility in the supply chain. If they trust in reducing their investment costs and what they need most is someone who is ready and willing to participate in a project, the exchange is to create a mechanism that guarantees that any person who enters is updated and has the opportunity to do so. opportunity to keep up. Otherwise, it is not sustainable as a labor market.
But companies are not doing that. They are not creating responsibility, trust or culture that would help phantom workers.
If you talk to any of these companies, most believe that we are going to automate this and we think: "I only need these people for a while". That is precisely our problem and that has historically been our problem. Since the industrial age: treating people who do contingent work badly that can not be automated. We stop paying attention to these people and their working conditions, we begin to treat them as something that can be replaced eventually, and we do not value the fact that they are doing something that a mechanical or computational process can not do.
I hate the parallel to horsepower. This is not like replacing horses with cars. People are not performing a mechanical task. They are extending something different about humans: their creativity and their interpretation.
What should we do to address this? What are the policy suggestions?
At a minimum, it means assessing everyone's contribution. The first step is to identify the people who have contributed. In Bangladesh, there was a big difference in textiles when companies that sold products had to tell us who was involved in making the shirt on my back. There must be a clear record thanking anyone who contributes labor to a product or service. The consumer should always be able to trace the supply chain of people who have had a hand to help them achieve their goals.
It is about regulating a form of employment that is not adjusted to full-time employment or to part-time employment or even clearly in self-employment. I think this is the time to say that job classification no longer works. Anyone who is of working age should have a baseline of provisions that are provided by the companies.
If companies want to use contract work with joy because they constantly need to share new ideas and new skills, the only way to make that good for both sides of the company is for people to jump into that group. And people do that when they have medical attention and other dispositions. This is the business case of universal medical care, of universal education as a public good. It will benefit every company.
I want to communicate with people who, in many ways, are describing working conditions. We are not describing a particular type of work. We are describing the current conditions for work based on project-based tasks. This can happen to everyone's jobs, and I hate that being the motivation because we should have cared all the time, since this has been happening to a lot of people. For me, the message of this book is: let's make it not only manageable, but also sustainable and pleasant. Stop making our lives involve work and start making work serve our lives.